In our previous newsletter here, we reported a decision from the Beijing Internet Court ruling that the copyright of a portrait generated by an artificial intelligence (“AI”) program is owned by the user who “controlled meticulously” the parameters for creating the image. This time, the Guangzhou Internet court had to tackle a different intellectual property issue relating to generative AI – liability for infringing work output from an AI program.
The plaintiff in this case was Shanghai Xinchuanghua Cultural Development Co., Ltd, being the exclusive licensee of the cartoon character Ultraman in mainland China. The plaintiff discovered that, upon inputting suitable textual prompts (e.g. draw an Ultraman) into an AI image generating platform operated by the defendant (whose name was redacted in the judgment), the platform was able to generate images that are highly similar to the authentic Ultraman character. The defendant argued that the AI system is provided by a third party, and the defendant only provided a platform for the public to access the AI system.
The court held that the AI generated character is substantially similar to Ultraman, with exact replication of certain unique features. The court also considered that copying of the Ultraman image is likely to have taken place, in view of the popularity of the Ultraman cartoon giving rise to a good chance that the AI system would have come across an original Ultraman image on the internet at some point. The court went on to find that the defendant was liable for copyright infringement on the basis that (i) the defendant is a generative AI service provider within the meaning of the PRC Interim Measures for Generative Artificial Intelligence Service Management (the “Interim Measures”), and (ii) the defendant failed to comply with the requirements under the Interim Measures in relation to establishing a complaint mechanism, provision of a warning to users not to infringe other’s intellectual property rights and the incorporation of appropriate marking to identify images created by an AI. As a result, the court ordered statutory damages of RMB 10,000 and the blocking of keywords on the defendant’s platform to prevent further generation of images that are substantially similar to Ultraman. However, the court did not support the plaintiff’s request to delete all Ultraman-related materials from the training data of the AI system on the basis that the defendant did not itself train the AI system.
While this appears on its face to be a simple case of copyright infringement with justice achieved in a relatively straightforward way, there are certain important questions relating to the interplay between AI and copyright that the court did not (or perhaps avoided) touching upon. For instance, although the developer of the AI system (the “Developer”) was not joined as a defendant in the case (bearing in mind that the defendant was merely a deployer of the AI system who made it available for public access), it would certainly have been worth the court discussing how liability for infringement ought to be apportioned between the Developer and the present defendant. Furthermore, in this case it appears that it is the Developer, rather than the defendant, who introduced images of Ultraman to the AI system, probably during the training stage. If so, then it would appear appropriate for the Developer to take a bigger portion of the blame for the copyright infringement. Also, would the Developer’s training of the AI system using an Ultraman image also constitute copyright infringement, or would the fair use exception apply? As to a defendant who provides the public with access to an AI system, it seems questionable whether the defendant was actually engaged in direct copying, though an infringement on the right to disseminate copyright works via the internet would likely stand (and the court did find that the defendant infringed this right).
This decision seems to beg more questions than it answers, and the court may have missed the opportunity to clarify certain highly controversial matters in the arena of AI and copyright. For now, AI developers and copyright practitioners will have to wait for another decision, probably from a higher court, or a judicial interpretation from the Supreme People’s Court, for further authoritative guidance on this topic.
(A version of this article was previously published by Managing Intellectual Property on 8 May 2024)